Targeted hydrolysis and decrepitation of Mn3AlC precipitates: a route to a novel precursor of rare earth free MnAl–C permanent magnets

In this work, Mn–Al–C alloys consisting of ϵ-MnAl and carbides of the types Mn3AlC and Al4C3 were produced and exposed to water at 80 °C for 35 d. This led to the hydrolysis and decrepitation of the carbides and ultimately, of the material. The hydrolysis behaviour was studied by combined mass spectrometry, thermogravimetry and differential scanning calorimetry. Microstructural studies with scanning electron microscopy and x-ray diffraction were carried out. X-ray computed tomography was used to study the evolution of cracks on a macroscopic scale. As the Mn3AlC precipitates were present as parallel platelets inside the ϵ-grains, the hydrolysis resulted in parallel voids separating regions of ϵ with the same orientation. This ϵ-skeleton could be easily milled into ϵ-flakes, which could be mechanically aligned so that their <001> directions were parallel. The ϵ-powder was transformed into the ferromagnetic τ-phase via heat treatment and is therefore suitable as a novel precursor material for the production of anisotropic τ-MnAl–C-based permanent magnets. Employing such a textured, ϵ-phase, precursor powder in an extrusion process could result in increased texture quality of the τ-phase with correspondingly enhanced magnetic properties.

(Some figures may appear in colour only in the online journal) * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Introduction
The aim of much recent research work is to develop new permanent magnet materials with a maximum energy product, (BH) max , in the gap between those of hard magnetic ferrites and sintered Nd-Fe-B i.e. 40 kJ m −3 < (BH) max ⩽ 400 kJ m −3 [1]. The reason for this is the critical nature of the rare earth elements [2] combined with the increasing demand for rare-earth-containing permanent magnets for the sustainable mobility and energy sectors [3,4]. Permanent magnet materials based on non-critical elements could be used to substitute rare-earth-containing magnets in certain applications [5]. With an estimated upper limit to (BH) max in excess of 100 kJ m −3 and containing no critical elements, τ -MnAl-C has very high potential for development as a rare-earth-free permanent magnet [6][7][8].
The current state of the art for τ -MnAl-C is materials with 40 kJ m −3 ⩽ (BH) max ⩽ 55 kJ m −3 produced by hot extrusion of an ε-MnAl-C precursor [9][10][11]. These (BH) max values are only approximately half of the estimated upper limit and the results were obtained using precursors consisting of the ε-MnAl-C phase (hP2, P63/mmc, Mg-type) in the form of solid polycrystalline billets [9], milled polycrystalline powder [10] or gas atomised spherical powder [11]. In all of these cases, no crystallographic alignment of the ε-phase grains is possible before the extrusion process. The transformation from ε to the ferromagnetic τ -phase (tP2, P4/mmm, AuCu-type) occurs during the hot extrusion process and although a crystallographic texture in the resulting τ -grains is produced via dynamic recrystallisation, the degree of alignment is not very high [10]. Starting from an ε precursor with a strong crystallographic alignment may act to improve the texture quality of the recrystallised τ -grains, thus improving the remanent magnetisation, and may also act to reduce the fraction nonrecrystallised regions in the microstructure, which result from poorly aligned ε-grains and are detrimental to the magnetic properties [12].
One possibility to introduce a crystallographic alignment into an ε precursor is if a powder could be produced which consists of single crystalline particles of the ε-phase having a flake morphology with a defined crystallographic orientation. The flakes would naturally tend to settle so that they are oriented parallel to each other, thus producing a strong overall crystallographic alignment. A possible route to producing such a powder can be envisaged when considering two results from the literature. The first result is that a microstructure consisting of ε-phase grains containing lamellar precipitates of Mn 3 AlC (cP5, Pm-3m, CaO 3 Ti-type) orientated parallel to the basal plane of the ε-phase can be produced in Mn-Al-C alloys using the so-called 'M' heat treatment [13,14]. From the starting state of a single ε-phase at high temperature (typically 1100 • C), the material is cooled to a lower temperature (typically between 800 • C and 900 • C) and, as the solubility limit of the ε-phase with respect to C decreases with decreasing temperature, precipitation of Mn 3 AlC occurs [13][14][15].
The second result of interest is that certain Mn-Al-C alloys tend to decrepitate after a certain period of time exposed  [15] showing the results of the decrepitation test carried out on Mn-Al-C alloys by Matsushita Electric Industrial Company (Data after Yamamoto [16]). The colour code shows the time in days that the material was exposed to air until first indications of decrepitation could be seen. Empty circles represent alloys that did not show any indications within the studied time period of 365 d. Reprinted from [15], Copyright (2017), with permission from Elsevier.
to air [16]. The dependence of the composition on the time until decrepitation is shown in figure 1, in which alloys from Yamamoto [16] are shown in a part of the isothermal section at 1100 • C of the ternary Mn-Al-C phase diagram after Bajenova et al [15]. Alloys that do not show signs of decrepitation are plotted as empty circles, while the decrepitated alloys are coloured according to the time until they decrepitated. Single phase fields are shown in dark grey, three-phase fields are shown in light grey. Both types of phase fields are labelled according to their phases. Two-phase fields are shown in white. As the samples used by Yamamoto were in an as-cast state it is unlikely that equilibrium was achieved and therefore, the compositions and volume fractions of the phases present in the alloys produced by Yamamoto may deviate from those suggested by the equilibrium diagram.
It can be seen from figure 1 that alloys with a composition located further inside a phase field containing Al 4 C 3 (hR7, R-3m, Al 4 C 3 -type) decrepitate faster. All the alloys that had decrepitated within 1 year contained Al 4 C 3 . This carbide is known to react with water to form solid Al hydroxide and gaseous methane (equation (1)), and this promotes the decrepitation of the material [16,17], Some of the alloys which had not decrepitated after a year contained Mn 3 AlC and therefore, the role of this compound in the decrepitation of Mn-Al-C alloys is less clear.
Studies in humid air and in water suggest a radical mechanism for the hydrolysis of Mn 3 AlC, as well as for other Mncarbides [18][19][20]. The hydrolysis of Mn 3 AlC can be described by equation (2). The Mn hydroxide and Al hydroxide products are solid at standard temperature and pressure and are insoluble in water but soluble in acids [21]; all the other products are gaseous [19]. The gaseous products consist mostly of hydrogen (74.7 vol.%) and methane (22.4 vol.%) and the remaining 2.9 vol.% is comprised of series of saturated (δ = 0) and a smaller fraction of unsaturated (δ > 0) higher hydrocarbons [19,22], The hydrolysis reaction proceeds readily in Mn 3 AlC and the linear corrosion rate was reported to be 0.3 µm h −1 [19].
In this work, various Mn-Al-C alloys with microstructures consisting of ε-phase grains containing lamellar precipitates of Mn 3 AlC will be produced and exposed to water at 80 • C for various durations up to 35 d. The aim of this work is to deliberately induce decrepitation of the material thus yielding a single-crystalline ε-phase powder with a defined flake geometry, which could act as a novel precursor for the production of MnAl-C magnets.

Methods
Ternary Mn-Al-C alloys with different compositions, namely Mn 53.7 Al 40.3 C 6, Mn 58 Al 35 C 7 and Mn 59 Al 31 C 10 were prepared by arc melting. The raw elements had a purity of 99.99% for the metallic elements Mn and Al, whereas C with 99.9% purity was used. The alloys were then homogenised at 1100 • C in Aratmosphere for 5 d and then quenched into water. This sample state is referred to as 'homogenised'. Parts of the homogenised Mn 53.7 Al 40.3 C 6 alloy were then subjected to the socalled M-treatment [13]: the material was reheated to 1100 • C and held for 20 min before furnace cooling to 850 • C and held for 20 min, and finally quenched into water. In addition to these alloys, an extruded magnet with the composition (Mn 54 Al 44 C 2 ) 99.4 Ni 0.6 was used in this work. The production of it is described in [10]. The composition of the alloys was checked by inductively coupled plasma optical emission spectroscopy and carrier-gas hot-extraction.
To study possible decrepitation of the sample material, bulk samples, whose surfaces had been prepared by grinding with SiC paper, were immersed in a water bath which was heated in an oven to 80 • C and held at that temperature for durations of up to 35 d. SiC with a grit size P600 was used to get comparable sample surfaces. The schematic set-up is shown in figure 2.
After decrepitation, parts of M-treated Mn 53.7 Al 40.3 C 6 were milled for 30 min in a vibration mill. The produced powder was spread on adhesive tape in preparation for x-ray diffraction (XRD)-measurements.
To analyse the microstructure of the samples, a Gemini Leo 1530 field emission gun scanning electron microscope (SEM) was used. The chemical composition was measured using energy dispersive x-ray spectrometry (EDX). This was done before and after the decrepitation experiments. As some samples decrepitated into powder, the material was embedded in conductive epoxy for better handling. Further structural characterisation was done by x-ray diffraction (XRD) with a Bruker diffractometer (Co-K α with λ(K α1 ) = 1.789 007 Å and λ(K α2 ) = 1.792 892 Å). This was done with sample material before and after the decrepitation experiment, similar to the SEM analysis. The resulting patterns were analysed using Rietveld refinement with FullProf Suite [23]. In order to analyse the difference of the measurements before and after the decrepitation experiment, the background was subtracted in the shown diffractograms. Raw data and the fit can be seen in supplementary 1.
The magnetic properties of the hot extruded magnet were measured before and after the decrepitation treatment with a permeameter by Magnet-Physik Dr Steingröver GmbH at room temperature. To ensure that the cylindrical samples with a diameter of 9.5 mm were saturated before the measurement, the samples were magnetised ex-situ using a 5 T field pulse before the measurement.
The magnetisation as a function on temperature, M(T), of powder samples was measured from 300 K to 900 K in a physical properties measurement system (PPMS) by Quantum design, using a vibrating sample magnetometer (VSM) with furnace option. The powder samples were mixed with a cement suitable for high temperature applications in order to mount them on the heating stick. A heating rate of 5 K min −1 and an applied magnetic field of µ 0 H = 0.1 T was used for the measurement.
X-ray computed tomography (XCT) was used to analyse the decrepitation behaviour of bulk samples after different durations in water at 80 • C on a macroscopic scale. For XCT, cubes of (Mn 54 Al 44 C 2 ) 99.4 Ni 0.6 and Mn 58 Al 35 C 7 of side 3 mm were prepared by spark erosion and their surface was prepared as described above. As the material is very brittle, it was not possible to obtain perfect cubes by this method. The scans were performed in a Phoenix Nanotom M device from General Electric. The samples were scanned using the same settings: a voltage of 125 kV and a current of 110 µA with an exposure time of 1500 ms. Due to the geometric setup, a voxel size of 2.375 µm was achieved. To reduce noise, five images were averaged for each of the 1080 angular positions and a 0.5 mm Al-filter was used. Furthermore, the cubes were mounted under a 45 • angle with respect to the XY-plane to reduce differences in the sample thickness when rotating the sample. This reduces so called Feldkamp-artefacts [24]. The 2D images were reconstructed with Phoenix datosjx2 software to a 3D image set and processed with the software VG Studio Max by Volume Graphics. The cubic samples were scanned before the start of the decrepitation experiment followed by scans every 7 d to gain insight about the crack growth.
A Setaram Sensys Eco device was used for differential scanning calorimetry (DSC) in combination with thermogravimetric analysis. The device was combined with an OmniStar quadrupole mass spectrometer from Pfeiffer Vacuum. Measurements were done in the initial state (homogenised or extruded) and after the decrepitation experiment to analyse the influence of the water bath and possible reaction products. After the decrepitation, the former bulk sample material cracked into coarse powder. To have more comparable conditions both the material in the starting state and the decrepitated material was hand crushed before the measurement. The sample material was heated in an Al 2 O 3 crucible from 30 • C to 500 • C with 5 K min −1 with an Ar-flow of 20 ml min −1 . The amount of analysed material was between 15 and 23 mg. Based on the mass for each measurement, the relative mass and the mass-related heat flow was used to compare the experiments within this work. As there are multiple mass to charge ratios (m/e) of interest, a spectrum of m/e from 0 u e −1 to 49 u e −1 was measured, resulting in 50 graphs per measurement. For reasons of clarity not all graphs can be shown in the corresponding diagrams and detailed explanation about how the data was used can be found in supplementary 2.

Results and discussion
The alloys produced in this work are shown in figure 3 as red squares on part of an isothermal section at 1100 • C of the ternary Mn-Al-C phase diagram after Bajenova et al [15]. After homogenisation at 1100 • C, it is expected from the diagram that the alloys Mn 53.7 Al 40.3 C 6 , Mn 58 Al 35 C 7 and Mn 59 Al 31 C 10 , will all consist of ε-MnAl and Mn 3 AlC and that an increasing volume fraction of Mn 3 AlC will be present with increasing C content.
Comparing figures 1 and 3, it is interesting to note that Mn 53.7 Al 40.3 C 6 lies on the border separating the decrepitated and intact alloys, studied by Yamamoto [16]. An extruded MnAl-based permanent magnet alloy with the composition (Mn 54 Al 44 C 2 ) 99.94 Ni 0.6 was also investigated here. The magnet composition was plotted in figure 3 by neglecting the Niaddition and only considering the Mn-Al-C ratio. It lies in the ε-MnAl phase field at 1100 • C, as intended.   [15]. The composition of the alloys in the present work are plotted as red squares. Reprinted from [15], Copyright (2017), with permission from Elsevier.
The microstructure of the homogenised state shows a grey matrix with an irregular distribution of dark precipitates with elongated shape. From the XRD pattern, the matrix can be identified as the ε-phase, however, no additional peaks corresponding to the dark precipitates are visible. The phase diagram (figure 3) suggests that this phase must be either Al 4 C 3 or Mn 3 AlC. For the M-treated sample state the microstructure additionally contains bright parallel platelets with different orientations in different matrix grains. The additional peaks visible in the XRD pattern were indexed as Mn 3 AlC. The formation of this special microstructure of ε-grains containing lamellar Mn 3 AlC precipitates is the aim of the M-treatment. EDX-measurements of the metallic elements were done to identify the dark appearing precipitates. They are considered to be the same phase for the homogenised and M-treated state (figures 4(a) and (b)) due to the same composition of the alloy and the same initial heat treatment step at 1100 • C. The area in which EDX spectra and maps were acquired is shown in figure 5(a) as a BSE image for M-treated Mn 53.7 Al 40.3 C 6 . Quantitative analysis of the EDX spectra can be found in table 1.
From the Mn-map (figure 5(b)) it can be seen that there is a strong difference in Mn-content between the ε-MnAl matrix and the precipitates that appear dark in the BSE-image ( figure 5(a)). The same areas that occur dark in the Mn-map show a high intensity in the Al-map (figure 5(c)), indicating the precipitates to be Al 4 C 3 . The point analysis from this carbide (table 1, Spot 5) gives an Al-content of almost 80 at.% and a Mn-content of around 20 at.%. The high measured Mn-content is likely due to the interaction volume overlapping onto neighbouring phases because of the small size of the precipitates. The point analysis confirms the much higher Al-content in the   dark precipitate compared to the ε-matrix (cf table 1, dark precipitate: Spot 5, ε-matrix: Spots 1 and 2). The bright appearing precipitates in figure 5(a) are visible as low intensity regions in the Al-map (figure 5(c)) and are shown to have a lower Alcontent than the matrix (cf table 1, bright precipitate: Spots 3 and 4, ε-matrix: Spots 1 and 2). The EDX measurements confirm that the bright, parallel, lamellar precipitates are Mn 3 AlC, in agreement with the XRD measurements. The dark precipitates are identified as Al 4 C 3 since there are no other binary Al-C carbides in the Mn-Al-C system and the alloy composition is very close the three-phase field of ε-MnAl, Mn 3 AlC and Al 4 C 3 in the 1100 • C isothermal section of the ternary phase diagram by Bajenova et al [15]. The microstructure of homogenised alloys Mn 58 Al 35 C 7 and Mn 59 Al 31 C 10 are shown in figure 6. For these nominal compositions, the C-content lies above the solubility limit of C in the ε-phase at 1100 • C and therefore lamellar Mn 3 AlC precipitates form during homogenisation and the M-treatment is not required. Comparing the C 7 alloy with the C 10 alloy, a higher volume fraction of plate-like Mn 3 AlC precipitates can be seen in C 13 , as expected. In addition, larger, blocky precipitates of Mn 3 AlC can be seen for both compositions (red arrows in figure 6). These might form during the solidification process when the local composition favours Mn 3 AlC nucleation prior to ε-MnAl.
The microstructure of the extruded magnet with composition (Mn 54 Al 44 C 2 ) 99.94 Ni 0.6 consists mainly of fine, recrystallised grains of the thermodynamically metastable τ -MnAl phase with small volume fractions of Mn 3 AlC and β-Mn precipitates with an equiaxed morphology, and is shown in detail by Feng et al [10]. The precipitates of Mn 3 AlC form at grain junctions where a locally high C-content is present due to the rejection of C to the growth front during the dynamic recrystallisation of the τ -MnAl grains.

Influence of Al 4 C 3 and Mn 3 AlC carbides.
In the homogenised state, the microstructure of the alloy with composition Mn 53.7 Al 40.3 C 6 consisted of the ε-phase and Al 4 C 3 .
In the M-treated state, lamellar Mn 3 AlC precipitates were also present. This gives an opportunity to study the different influences of the Al 4 C 3 and Mn 3 AlC carbides on the decrepitation of the material. In figure 7, BSE images of both samples are shown after the decrepitation experiment (a and c). EDX maps were acquired in the red square for Mn, Al and C ((b) and (d)). Regions of high intensity in the EDX C-map  correspond to the epoxy resin in which the material was embedded for microscopy. From figure 7 it is apparent that the hydrolysis and decrepitation of Al 4 C 3 and Mn 3 AlC is at a very advanced state: only a very few of the lamellar Mn 3 AlC precipitates, which appear bright in the BSE image, remain (red arrows in figure 5(c)) and a few remaining Al 4 C 3 grains can be seen from the bright regions in the Al EDX maps corresponding to dark regions in the Mn EDX maps. Instead, a large volume fraction of voids is present which were previously occupied by the carbides. The hydrolysis of both carbides leads to gaseous and solid reaction products (equations (1) and (2)) and although the solid products (Mn hydroxide and Al hydroxide) are insoluble in water, they appear to have been carried away from the initial location of the carbide, thus creating the voids. Brown precipitates could be seen in the petri dish during the experiment which is expected to be a manganese oxide-hydroxide mixture which was also reported by Karen et al [19]. The lamellar Mn 3 AlC precipitates initially spanned the entire width of the ε-grains, which are of the order of 100 µm in size. Assuming that the reported linear corrosion rate for Mn 3 AlC in water at 60 • C (0.3 µm h −1 [19]) is similar to that at 80 • C and does not change over the course of the hydrolysis reaction, after 35 d (840 h), the distance over which the corrosion had occurred would be of the same order of magnitude as the size of the ε-grains. It is therefore expected that the Mn 3 AlC precipitates would be completely hydrolysed and decrepitated under the experimental conditions. The few remaining Mn 3 AlC precipitates may have been protected from contact with the water by other grains thus experiencing a lower corrosion rate.
In figure 7(c), the ε-grains are separated by large intergranular cracks. Within each grain, several lamellar, parallel voids are present where the Mn 3 AlC precipitates were located and in addition, finer cracks are visible running in all directions. A network of fine cracks is also visible in figure 7(a). Both the large intergranular and fine intragranular cracks in both sample states are likely to result from stresses generated during the hydrolysis of the carbides.

Structural stability of C-rich Mn-Al-C alloys.
As shown above, Mn 3 AlC carbides can lead to cracking on a microscopic level. In addition, macroscopic studies were done by studying the decrepitation process of a cubic sample of Mn 58 Al 35 C 7 using XCT. Therefore, the cube was scanned by the XCT in an interval of 7 d during the decrepitation experiment. The results are shown in figure 8 as a 3D reconstruction and a 2D slice of the sample volume.
Already at the beginning of the experiment, shown in figure 8(a), the corners of the cube were broken off due to the brittle character of the material. In the starting state, small cracks, which appear bright in the 3D reconstruction, are present. These are likely to be due to thermal stresses formed during the quenching of the material after homogenisation at 1100 • C. After 7 d, these existing cracks grew and additional cracks formed in the surrounding area. In the 3D reconstruction a white cloud can be seen. These are inner surfaces of cracks detected by the analysis software. With increasing duration in water at 80 • C, the cracks grow and new cracks from different directions impinge on each other (figure 8(c)). After 21 d, a parallel crack mesh is visible on the macroscopic level and some parts of the former cube had become completely separated. As the sample was very brittle at this stage, no further XCT scans were carried out in order to avoid damaging the sample through handling it and thus influence the cracking behaviour. The sample was retained in hot water until a duration of 35 d had been reached to ensure that the microstructure was comparable with other samples studied in this work.
The homogenised state of Mn 58 Al 35 C 7 is shown in figure 6(a) and for comparison, a BSE image taken after the decrepitation experiment is shown in figure 9(a). An enlarged image of a cracked area (red rectangle) can be seen in figure 9(b). For SEM analysis, the material was embedded in epoxy resin as it has decrepitated into a coarse powder. The overview image shows parallel voids within one grain of ε-MnAl, similar to figure 7(c); however, almost no fine cracks not parallel to the main direction can be seen, in contrast to the microstructure of the decrepitated, M-treated Mn 53.7 Al 40.3 C 6 alloy. This is likely to be connected to the absence of randomly  distributed Al 4 C 3 precipitates in the microstructure of the Mn 58 Al 35 C 7 . The enlarged area in figure 9(b) shows the parallelism between the cracks and some remaining Mn 3 AlC that had not reacted completely during the decrepitation experiment (red arrow). A diffractogram of the starting state for both Mn 58 Al 35 C 7 and Mn 59 Al 31 C 10 was taken from crushed sample material to make the measurement comparable with the measurement after the decrepitation experiment, which was done using decrepitated, crushed material. The diffractograms of both states were analysed using Rietveld refinement. The mass fractions of the phases extracted from this data are shown in table 2. It can be seen that the Mn 3 AlC phase fraction is higher before the decrepitation experiment. This indicates a loss of Mn 3 AlC which is due to the hydrolysis (equation (2)) during the experiment, resulting in the decrepitation of the material [19].
In order to study the hydrolysis of Mn 3 AlC in the various different materials, mass spectroscopy of the decrepitated material was carried out whilst continuously heating the sample from 30 • C to 500 • C. Several mass/charge (m/e) lines were excluded from the analysis, as they showed a constant behaviour over time or a low signal in general. Furthermore, high ion currents for one m/e line can lead to an overshadowing of neighbouring m/e lines, which was seen for Ar-based ions as well as for N 2 -based ions. In addition, a number of ratios were not considered because they are based on ions of the residual atmosphere (e.g. N 2 + ) or the Ar-atmosphere during the experiment. A more detailed overview about the selection of m/e lines is given in supplementary 2.
In figure 10 the selected m/e traces are shown as solid lines; the colour key is shown below the diagrams. For the red and purple m/e lines, the combined signal for more than one ion is shown, as different ions have the same mass to charge ratio and therefore no deconvolution of the signal for each ion type can be achieved. For better comparison measurements were done before (figures 10(a)-(c)) and after (figures 10(d)-(f)) the decrepitation experiments for Mn 58 Al 35 C 7 (figure 10, left column), Mn 59 Al 31 C 10 (figure 10, right column) and the extruded (Mn 53 Al 45 C 2 ) 99.4 Ni 0.6 magnet (figure 10, right column). Complementary to the mass spectrometer data, the relative mass acquired simultaneously from thermogravimetry (TG) data (black dotted lines) and related heat flow from DSC data (red dotted lines) are also shown for each of the six experiments.
The m/e lines for figures 10(a)-(c) show similar behaviour. For all the three experiments the ion signals show a decreasing trend, which is typical for the current experimental equipment. The time in which the measurement system is under Ar-flow increases during the experiment and therefore all ion signals not related to Ar decrease to a minimum, unless a reaction in the sample chamber takes place. The red line (OH + and NH 3 + ) shows a slightly different decreasing behaviour, which may be related to some degassing effects from the crucible or powder. The TG signal for all three alloys shows a slight increase of about 0.5%, which seems to be an experimental artefact, that also occurs in control measurements with an empty crucible (see supplementary 4). The heat flow (DSC) measurements show only minor deviation from the control measurement. In conclusion, in the starting state, the behaviour of all the materials is similar and no reaction or evaporation takes place for the sample state before decrepitation.
In contrast, large differences in the curves of the various samples can be seen after the decrepitation experiments. For both C-rich samples a significant mass loss can be seen during the heating. This feature of the TG signal goes along with a change of ion current of different water-based species (H 2 O + , H 3 O + , OH + ) but also a broad peak in the ion current of some organic species (shared line for C 3 H 8 + , CO 2 + and C 2 H 5 O + ) can be seen. In addition, an endothermic peak in the heat flow signal is seen for both C-rich alloys. The magnet composition does not show these effects and in the case of that composition, the curves before and after the decrepitation are similar ( figure 10, right column). At this point, only the C-rich alloys will be compared, as the stability of the MnAl-magnet will be discussed further below.
For decrepitated Mn 58 Al 35 C 7 roughly 4% of the initial mass is lost after the DSC-TG-MS experiment. The loss occurs over a broad temperature range and two peaks in the ion current of water-related ion species as well as for organic species can be seen at around 200 • C and 275 • C. This might indicate different kinds of reaction products that are evaporating or other origins of the reaction product that are activated at various temperatures.
Mn 59 Al 31 C 10 exhibited a higher mass loss than Mn 58 Al 35 C 7 (cf 8% for Mn 59 Al 31 C 10 , 4% for Mn 58 Al 35 C 7 ) which is attributed to the higher volume fraction of Mn 3 AlC in Mn 59 Al 31 C 10 . Accordingly, also larger peaks in the ion current and heat flow can be seen for Mn 59 Al 31 C 10 . Compared to Mn 58 Al 35 C 7 , the peak for the m/e lines for Mn 59 Al 31 C 10 is higher and sharper as it occurs between 225 • C and 275 • C for the water related species (OH + , H 2 O + , H 3 O + ). Furthermore, only one peak can be seen, unlike for Mn 58 Al 35 C 7 . The peaks for the remaining species (C 3 H 6 + and C 2 H 5 O + ) appear weaker.
As the hydrolysis reaction products in both alloy compositions are the same it is yet unclear how these differences between Mn 58 Al 35 C 7 and Mn 59 Al 31 C 10 can be interpreted. There are several possibilities like dissociation of hydroxides or possible water reservoirs within the voids and cracks of the decrepitated samples that need to be considered as origin for the MS signals.

Resistance to decrepitation of extruded MnAl-C-based magnet.
A cube of an extruded magnet with the composition (Mn 54 Al 44 C 2 ) 99.4 Ni 0.6 was exposed to water at 80 • C and was scanned by XCT. The reconstructed 3D image as well as a 2D slice through the cube is shown in figure 11 at the beginning and the end of the decrepitation experiment.
The spark-eroded cube shows some imperfections in the starting state as mentioned above. In addition, a small number of cracks is present in the initial state (red marked area). This gives a comparable starting state to the Mn 58 Al 35 C 7 cube shown in figure 8, as both cubes show existing cracks in the beginning of the experiment. However, for (Mn 53 Al 45 C 2 ) 99.4 Ni 0.6 no crack growth with increasing duration in hot water was observed and no additional cracks formed. Even after 35 d the cube looks unchanged in the XCT images ( figure 11(b)). This result, combined with the fact that the TG-MS-DSC data showed no difference in the starting and final states (figures 10(c) and (f), respectively) demonstrates the excellent structural stability of extruded (Mn 53 Al 45 C 2 ) 99.4 Ni 0.6 magnet in water at 80 • C, despite the presence of a small volume fraction of fine, equiaxed Mn 3 AlC precipitates.
In addition to showing the structural integrity and chemical stability of the typical magnet composition, it is important to investigate whether a degradation of the magnetic properties occurred during the immersion in hot water. A cylindershaped sample of the magnet was first magnetically saturated using a pulsed magnetic field and was then measured in a permeameter before exposing it to hot water at 80 • C for 35 d. After the immersion in hot water, the sample was re-saturated and remeasured. The demagnetising branch of the polarisation curves is shown in figure 12.
Before and after the decrepitation experiment the characteristic values for remanence and coercivity were at 0.498 T and 0.502 T, 268 kA m −1 and 265 kA m −1 , respectively. The differences in remanence and coercivity are within the error of the measurement. The maximum of the energy product (BH) max was at 38 kJ m −3 for both states, showing that no degradation of the magnetic properties happened during the exposure to water at 80 • C for 35 d.

Synthesis of a novel single crystalline epsilon phase powder
After decrepitation, the microstructure of M-treated Mn 53.7 Al 40.3 C 6 alloy consists of ε-MnAl with parallel voids where the Mn 3 AlC precipitates were previously located, and a network of coarse intergranular and fine intragranular cracks (figures 7(c) and (d)). As shown by Kubo et al [13], the Mtreatment leads to the precipitation of Mn 3 AlC with a defined orientation relationship with ε-MnAl: ε (0001) ∥ Mn 3 AlC (111). The voids left by the Mn 3 AlC precipitates after hydrolysis and decrepitation are therefore also parallel to the basal plane of the ε-phase. The voids and intergranular cracks are structural weak points and mechanical comminution of the remaining ε-phase skeleton should result in a single crystalline powder with a flake morphology, where the basal plane of the crystal structure corresponds to the plane of the flake. In order to investigate this, the decrepitated material was hand crushed and milled as described above. In figure 13 a secondary electron image of the material after milling (a) and powder XRD patterns for different material states are shown (b).
After milling the decrepitated material, flakes of Mn 53.7 Al 40.3 C 6 in various sizes can be seen. The flake morphology means that the particles should align well with their extended plane parallel, e.g. when being spread on an XRD sample holder. X-ray diffractograms for the different sample states are shown in figure 6(b). Similar to figure 3(c), the XRD after M-treatment shows both, ε-MnAl and Mn 3 AlC peaks from a variety of lattice planes. After the decrepitation experiments the Mn 3 AlC peaks are missing for both the crushed and milled material, which is expected due the hydrolysis and decrepitation of the carbide; however, many peaks of ε-MnAl have also vanished or decreased significantly in intensity. This is even more distinctive after milling where only the (0002) and the ( 1012 ) peak of ε-MnAl can be seen. The difference between the crushed and milled states can be attributed to a more complete separation of the ε-phase skeleton into flake particles by the milling process.
These results indicate that the milled powder after decrepitation consists of ε-MnAl flakes with their thin axis aligned to the <001> direction. As each ε-grain is intersected by multiple Mn 3 AlC plates (see figure 4(b)) this method results in single crystalline ε-MnAl microflakes with a defined crystallographic orientation.
In order to demonstrate the possibility of transforming the novel ε-powder into the ferromagnetic τ -phase, which is required for permanent magnet applications, the milled powder sample was heated from 300 K to 900 K in an applied magnetic field of µ 0 H = 0.1 T and the magnetisation, M, was recorded as a function of the temperature. The resulting M(T) curve and its first derivative, dM/dT(T), are shown in figure 14. In the temperature range 300 K < T < 315 K, which corresponds to reported values of the Curie temperature of Mn 3 AlC [18,25], there is a small decrease in magnetisation on the heating branch and a corresponding increase on the cooling branch. These effects are attributed to the presence of a small volume fraction of Mn 3 AlC remaining in the material, which was too small to be detected by XRD (see figure 13(b)). At T > 315 K on the heating branch, the magnetisation stays at low levels. This is expected as the material consists almost entirely of the ε-phase, which is paramagnetic at these temperatures [26,27]. The ε → τ phase transformation typically occurs in the temperature range 573 K < T < 1023 K [14,28,29] and the large increase in the magnetisation visible on the cooling branch indicates that this transformation had occurred during the experiment. This is further confirmed by the fact that the pronounced minimum of the first derivative curve, which represents the Curie temperature, was determined to be 520 K, which is similar to reported values of the Curie temperature of τ -MnAl-C [30]. As the novel powder consisted initially of single crystalline flakes of the ε-phase, the ε → τ phase transformation likely to have proceeded via the displacivediffusional mode [31,32] rather than the massive mode [33,34], which requires the presence of ε-ε grain boundaries.

Conclusions
The results of the current work show that it is possible to obtain a single-crystalline, ε-phase powder with a flake particle morphology via hydrolysis and decrepitation of Mn 3 AlC precipitates. The initial microstructure of the alloys studied here consisted of ε-grains containing many lamellar Mn 3 AlC precipitates which spanned the entire width of the grains and were oriented with their short axis parallel to the <001> direction of ε. The hydrolysis and decrepitation of Mn 3 AlC led to the almost complete removal of that phase from the microstructure after 35 d exposed to water at 80 • C, leaving a skeleton of ε-grains. After gentle milling, single crystalline flakes of ε were obtained, where the <001> direction was parallel to the short axis of the flakes. The shape of the flakes did not result from deformation during milling but was already present in the microstructure. The single crystalline flakes were therefore not expected to contain a significant fraction of crystalline defects. A strong, <001> texture was produced by mechanically aligning the flakes. The ε-phase powder was successfully transformed into the ferromagnetic τ -phase via heat treatment, demonstrating its suitability as a novel precursor for MnAl-C-based, rare-earth-free permanent magnets. As a textured ε precursor reduces the number of possible crystallographic orientations for the hard magnetic τ -phase, a higher texture quality in extruded permanent magnets may result; however, further studies are required to assess both this and, subsequently, whether the proposed method is suitable for large-scale production.
A MnAl-C-Ni magnet, which contained a small volume fraction of equiaxed Mn 3 AlC precipitates at the junctions of the τ -phase grains, was exposed to the same decrepitation experiment. The magnet showed no signs of cracking after 35 d, as well as no degradation of its magnetic properties. This indicates the excellent resistance to corrosion of this material in water at elevated temperatures.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).